Physics-Informed Machine Learning for Real-time Reservoir Management
summary
A physics-informed ML workflow combines a site behavior library built from multi-fidelity simulations (PFlowTran high‑fidelity runs plus reduced‑order Patsak models), lab and MSEEL field data, and ML inverse/forward models to infer ~5 key site parameters and forecast cumulative and stage production and spatial saturation. The training varies ~13–15 input parameters, achieves R2 > 0.90 on test cases, and yields ML forward models that are 10^4–10^7× faster than full simulations while qualitatively matching observed MSEEL production.
AI-written summary · about
chapters (9)
- 0:00Introduction and overview
- 4:14MC vapes Enfield site details
- 7:45Pressure management premise
- 11:19Proposed ML workflow
- 16:23Field and lab scale data
- 21:44High-fidelity library development
- 23:32Preliminary results
- 27:29Next steps and conclusions
- 27:50Q&A session
related talks
24mUnder Pressure: Applying ML in Real Time
44mML in Production
36mThe ML Monitoring Flow for Models Deployed to Production
31mMLflow: Platform for Complete Machine Learning Lifecycle
32mAutomating Production Production Level Machine Learning on AWS
16mTransforming Inventory Forecasting with Production ML
30mReconceptualizing Machine Learning for the Real-time World
33mProduction ML Models in the Mobile Gaming Industry